Closed DanielDoehring closed 1 year ago
Sounds reasonable to me, maybe we could move these to a separate set of allocation tests. Are there downsides you can think of (i.e., runtime, inconsistent failures, etc)?
Runtime should not increase significantly (unless we move them to own tests, as argued by Hendrik below), as (IIRC) 75%+ test time is spent for compilation.
Yes, this would be nice!
Sounds reasonable to me, maybe we could move these to a separate set of allocation tests. Are there downsides you can think of (i.e., runtime, inconsistent failures, etc)?
Moving them to a separate test job would likely have a significant impact on total test time since most time is spent in compilation. If we just include the stuff directly after testing an elixir, everything should be compiled already and the impact on total CI time should be small.
Ok, I will start with the non-threaded tests and see how this evolves.
Added tests (currently passing):
test_dgmulti_1d.jl
test_dgmulti_2d.jl
test_dgmulti_3d.jl
test_mpi_p4est_2d.jl
test_mpi_p4est_3d.jl
test_mpi_tree.jl
test_p4est_2d.jl
test_p4est_3d.jl
test_paper_self_gravitating_gas_dynamics.jl
test_parabolic_1d.jl
test_parabolic_2d.jl
test_parabolic_3d.jl
test_structured_1d.jl
test_structured_2d.jl
test_structured_3d.jl
test_t8code_2d.jl
test_threaded.jl
test_tree_1d_advection.jl
test_tree_1d_burgers.jl
test_tree_1d_euler.jl
test_tree_1d_eulergravity.jl
test_tree_1d_eulermulti.jl
test_tree_1d_fdsbp.jl
test_tree_1d_hypdiff.jl
test_tree_1d_mhd.jl
test_tree_1d_mhdmulti.jl
test_tree_1d_shallowwater_twolayer.jl
test_tree_1d_shallowwater.jl
test_tree_2d_acoustics.jl
test_tree_2d_advection.jl
test_tree_2d_euler.jl
test_tree_2d_euleracoustics.jl
test_tree_2d_eulermulti.jl
test_tree_2d_fdsbp.jl
test_tree_2d_hypdiff.jl
test_tree_2d_kpp.jl
test_tree_2d_lbm.jl
test_tree_2d_linearizedeuler.jl
test_tree_2d_mhd.jl
test_tree_2d_mhdmulti.jl
test_tree_2d_shallowwater_twolayer.jl
test_tree_2d_shallowwater.jl
test_tree_3d_advection.jl
test_tree_3d_euler.jl
test_tree_3d_eulergravity.jl
test_tree_3d_fdsbp.jl
test_tree_3d_hypdiff.jl
test_tree_3d_lbm.jl
test_tree_3d_mhd.jl
test_unstructured_2d.jl
Recently, we observed allocations due to upstream/Julia changes for a variety of cases, see e.g.
1656 , #1642, #1635 #1626.
We test for a variety of cases whether the
rhs!
allocates, see for instancehttps://github.com/trixi-framework/Trixi.jl/blob/22856f4af27a06d38935d4f373a46f3492a4bf08/test/test_p4est_2d.jl#L27-L34
Any opinions on including this for every/more tests to spot allocations faster?